Linux Audio

Check our new training course

Loading...
v6.9.4
  1/* SPDX-License-Identifier: GPL-2.0 */
  2/* Copyright(c) 2013 - 2018 Intel Corporation. */
  3
  4#ifndef _I40E_TXRX_H_
  5#define _I40E_TXRX_H_
  6
  7#include <net/xdp.h>
  8#include "i40e_type.h"
  9
 10/* Interrupt Throttling and Rate Limiting Goodies */
 11#define I40E_DEFAULT_IRQ_WORK      256
 12
 13/* The datasheet for the X710 and XL710 indicate that the maximum value for
 14 * the ITR is 8160usec which is then called out as 0xFF0 with a 2usec
 15 * resolution. 8160 is 0x1FE0 when written out in hex. So instead of storing
 16 * the register value which is divided by 2 lets use the actual values and
 17 * avoid an excessive amount of translation.
 18 */
 19#define I40E_ITR_DYNAMIC	0x8000	/* use top bit as a flag */
 20#define I40E_ITR_MASK		0x1FFE	/* mask for ITR register value */
 21#define I40E_MIN_ITR		     2	/* reg uses 2 usec resolution */
 
 
 22#define I40E_ITR_20K		    50
 
 23#define I40E_ITR_8K		   122
 24#define I40E_MAX_ITR		  8160	/* maximum value as per datasheet */
 25#define ITR_TO_REG(setting) ((setting) & ~I40E_ITR_DYNAMIC)
 26#define ITR_REG_ALIGN(setting) __ALIGN_MASK(setting, ~I40E_ITR_MASK)
 27#define ITR_IS_DYNAMIC(setting) (!!((setting) & I40E_ITR_DYNAMIC))
 28
 29#define I40E_ITR_RX_DEF		(I40E_ITR_20K | I40E_ITR_DYNAMIC)
 30#define I40E_ITR_TX_DEF		(I40E_ITR_20K | I40E_ITR_DYNAMIC)
 31
 32/* 0x40 is the enable bit for interrupt rate limiting, and must be set if
 33 * the value of the rate limit is non-zero
 34 */
 35#define INTRL_ENA                  BIT(6)
 36#define I40E_MAX_INTRL             0x3B    /* reg uses 4 usec resolution */
 37#define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
 38
 39/**
 40 * i40e_intrl_usec_to_reg - convert interrupt rate limit to register
 41 * @intrl: interrupt rate limit to convert
 42 *
 43 * This function converts a decimal interrupt rate limit to the appropriate
 44 * register format expected by the firmware when setting interrupt rate limit.
 45 */
 46static inline u16 i40e_intrl_usec_to_reg(int intrl)
 47{
 48	if (intrl >> 2)
 49		return ((intrl >> 2) | INTRL_ENA);
 50	else
 51		return 0;
 52}
 
 
 
 53
 54#define I40E_QUEUE_END_OF_LIST 0x7FF
 55
 56/* this enum matches hardware bits and is meant to be used by DYN_CTLN
 57 * registers and QINT registers or more generally anywhere in the manual
 58 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
 59 * register but instead is a special value meaning "don't update" ITR0/1/2.
 60 */
 61enum i40e_dyn_idx {
 62	I40E_IDX_ITR0 = 0,
 63	I40E_IDX_ITR1 = 1,
 64	I40E_IDX_ITR2 = 2,
 65	I40E_ITR_NONE = 3	/* ITR_NONE must not be used as an index */
 66};
 67
 68/* these are indexes into ITRN registers */
 69#define I40E_RX_ITR    I40E_IDX_ITR0
 70#define I40E_TX_ITR    I40E_IDX_ITR1
 71#define I40E_SW_ITR    I40E_IDX_ITR2
 72
 73/* Supported RSS offloads */
 74#define I40E_DEFAULT_RSS_HENA ( \
 75	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
 76	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
 77	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
 78	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
 79	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
 80	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
 81	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
 82	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
 83	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
 84	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
 85	BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
 86
 87#define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
 88	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
 89	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
 90	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
 91	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
 92	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
 93	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
 94
 95#define i40e_pf_get_default_rss_hena(pf) \
 96	(test_bit(I40E_HW_CAP_MULTI_TCP_UDP_RSS_PCTYPE, (pf)->hw.caps) ? \
 97	 I40E_DEFAULT_RSS_HENA_EXPANDED : I40E_DEFAULT_RSS_HENA)
 98
 99/* Supported Rx Buffer Sizes (a multiple of 128) */
100#define I40E_RXBUFFER_256   256
101#define I40E_RXBUFFER_1536  1536  /* 128B aligned standard Ethernet frame */
102#define I40E_RXBUFFER_2048  2048
103#define I40E_RXBUFFER_3072  3072  /* Used for large frames w/ padding */
104#define I40E_MAX_RXBUFFER   9728  /* largest size for single descriptor */
105
106/* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
107 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
108 * this adds up to 512 bytes of extra data meaning the smallest allocation
109 * we could have is 1K.
110 * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
111 * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
112 */
113#define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
114#define I40E_PACKET_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
115#define i40e_rx_desc i40e_16byte_rx_desc
116
117#define I40E_RX_DMA_ATTR \
118	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
119
120/* Attempt to maximize the headroom available for incoming frames.  We
121 * use a 2K buffer for receives and need 1536/1534 to store the data for
122 * the frame.  This leaves us with 512 bytes of room.  From that we need
123 * to deduct the space needed for the shared info and the padding needed
124 * to IP align the frame.
125 *
126 * Note: For cache line sizes 256 or larger this value is going to end
127 *	 up negative.  In these cases we should fall back to the legacy
128 *	 receive path.
129 */
130#if (PAGE_SIZE < 8192)
131#define I40E_2K_TOO_SMALL_WITH_PADDING \
132((NET_SKB_PAD + I40E_RXBUFFER_1536) > SKB_WITH_OVERHEAD(I40E_RXBUFFER_2048))
133
134static inline int i40e_compute_pad(int rx_buf_len)
135{
136	int page_size, pad_size;
137
138	page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
139	pad_size = SKB_WITH_OVERHEAD(page_size) - rx_buf_len;
140
141	return pad_size;
142}
143
144static inline int i40e_skb_pad(void)
145{
146	int rx_buf_len;
147
148	/* If a 2K buffer cannot handle a standard Ethernet frame then
149	 * optimize padding for a 3K buffer instead of a 1.5K buffer.
150	 *
151	 * For a 3K buffer we need to add enough padding to allow for
152	 * tailroom due to NET_IP_ALIGN possibly shifting us out of
153	 * cache-line alignment.
154	 */
155	if (I40E_2K_TOO_SMALL_WITH_PADDING)
156		rx_buf_len = I40E_RXBUFFER_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
157	else
158		rx_buf_len = I40E_RXBUFFER_1536;
159
160	/* if needed make room for NET_IP_ALIGN */
161	rx_buf_len -= NET_IP_ALIGN;
162
163	return i40e_compute_pad(rx_buf_len);
164}
165
166#define I40E_SKB_PAD i40e_skb_pad()
167#else
168#define I40E_2K_TOO_SMALL_WITH_PADDING false
169#define I40E_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
170#endif
171
172/**
173 * i40e_test_staterr - tests bits in Rx descriptor status and error fields
174 * @rx_desc: pointer to receive descriptor (in le64 format)
175 * @stat_err_bits: value to mask
176 *
177 * This function does some fast chicanery in order to return the
178 * value of the mask which is really only used for boolean tests.
179 * The status_error_len doesn't need to be shifted because it begins
180 * at offset zero.
181 */
182static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
183				     const u64 stat_err_bits)
184{
185	return !!(rx_desc->wb.qword1.status_error_len &
186		  cpu_to_le64(stat_err_bits));
187}
188
189/* How many Rx Buffers do we bundle into one write to the hardware ? */
190#define I40E_RX_BUFFER_WRITE	32	/* Must be power of 2 */
 
 
 
 
 
 
 
191
192#define I40E_RX_NEXT_DESC(r, i, n)		\
193	do {					\
194		(i)++;				\
195		if ((i) == (r)->count)		\
196			i = 0;			\
197		(n) = I40E_RX_DESC((r), (i));	\
198	} while (0)
199
 
 
 
 
 
200
201#define I40E_MAX_BUFFER_TXD	8
202#define I40E_MIN_TX_LEN		17
203
204/* The size limit for a transmit buffer in a descriptor is (16K - 1).
205 * In order to align with the read requests we will align the value to
206 * the nearest 4K which represents our maximum read request size.
207 */
208#define I40E_MAX_READ_REQ_SIZE		4096
209#define I40E_MAX_DATA_PER_TXD		(16 * 1024 - 1)
210#define I40E_MAX_DATA_PER_TXD_ALIGNED \
211	(I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
212
213/**
214 * i40e_txd_use_count  - estimate the number of descriptors needed for Tx
215 * @size: transmit request size in bytes
216 *
217 * Due to hardware alignment restrictions (4K alignment), we need to
218 * assume that we can have no more than 12K of data per descriptor, even
219 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
220 * Thus, we need to divide by 12K. But division is slow! Instead,
221 * we decompose the operation into shifts and one relatively cheap
222 * multiply operation.
223 *
224 * To divide by 12K, we first divide by 4K, then divide by 3:
225 *     To divide by 4K, shift right by 12 bits
226 *     To divide by 3, multiply by 85, then divide by 256
227 *     (Divide by 256 is done by shifting right by 8 bits)
228 * Finally, we add one to round up. Because 256 isn't an exact multiple of
229 * 3, we'll underestimate near each multiple of 12K. This is actually more
230 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
231 * segment.  For our purposes this is accurate out to 1M which is orders of
232 * magnitude greater than our largest possible GSO size.
233 *
234 * This would then be implemented as:
235 *     return (((size >> 12) * 85) >> 8) + 1;
236 *
237 * Since multiplication and division are commutative, we can reorder
238 * operations into:
239 *     return ((size * 85) >> 20) + 1;
240 */
241static inline unsigned int i40e_txd_use_count(unsigned int size)
242{
243	return ((size * 85) >> 20) + 1;
244}
245
246/* Tx Descriptors needed, worst case */
247#define DESC_NEEDED (MAX_SKB_FRAGS + 6)
 
248
249#define I40E_TX_FLAGS_HW_VLAN		BIT(1)
250#define I40E_TX_FLAGS_SW_VLAN		BIT(2)
251#define I40E_TX_FLAGS_TSO		BIT(3)
252#define I40E_TX_FLAGS_IPV4		BIT(4)
253#define I40E_TX_FLAGS_IPV6		BIT(5)
 
 
254#define I40E_TX_FLAGS_TSYN		BIT(8)
255#define I40E_TX_FLAGS_FD_SB		BIT(9)
256#define I40E_TX_FLAGS_UDP_TUNNEL	BIT(10)
257#define I40E_TX_FLAGS_VLAN_MASK		0xffff0000
258#define I40E_TX_FLAGS_VLAN_PRIO_MASK	0xe0000000
259#define I40E_TX_FLAGS_VLAN_PRIO_SHIFT	29
260#define I40E_TX_FLAGS_VLAN_SHIFT	16
261
262struct i40e_tx_buffer {
263	struct i40e_tx_desc *next_to_watch;
264	union {
265		struct xdp_frame *xdpf;
266		struct sk_buff *skb;
267		void *raw_buf;
268	};
269	unsigned int bytecount;
270	unsigned short gso_segs;
271
272	DEFINE_DMA_UNMAP_ADDR(dma);
273	DEFINE_DMA_UNMAP_LEN(len);
274	u32 tx_flags;
275};
276
277struct i40e_rx_buffer {
278	dma_addr_t dma;
279	struct page *page;
280	__u32 page_offset;
281	__u16 pagecnt_bias;
282	__u32 page_count;
 
 
 
 
 
 
 
283};
284
285struct i40e_queue_stats {
286	u64 packets;
287	u64 bytes;
288};
289
290struct i40e_tx_queue_stats {
291	u64 restart_queue;
292	u64 tx_busy;
293	u64 tx_done_old;
294	u64 tx_linearize;
295	u64 tx_force_wb;
296	u64 tx_stopped;
297	int prev_pkt_ctr;
298};
299
300struct i40e_rx_queue_stats {
301	u64 non_eop_descs;
302	u64 alloc_page_failed;
303	u64 alloc_buff_failed;
304	u64 page_reuse_count;
305	u64 page_alloc_count;
306	u64 page_waive_count;
307	u64 page_busy_count;
308};
309
310enum i40e_ring_state {
311	__I40E_TX_FDIR_INIT_DONE,
312	__I40E_TX_XPS_INIT_DONE,
313	__I40E_RING_STATE_NBITS /* must be last */
314};
315
316/* some useful defines for virtchannel interface, which
317 * is the only remaining user of header split
318 */
 
319#define I40E_RX_DTYPE_HEADER_SPLIT  1
 
320#define I40E_RX_SPLIT_L2      0x1
321#define I40E_RX_SPLIT_IP      0x2
322#define I40E_RX_SPLIT_TCP_UDP 0x4
323#define I40E_RX_SPLIT_SCTP    0x8
324
325/* struct that defines a descriptor ring, associated with a VSI */
326struct i40e_ring {
327	struct i40e_ring *next;		/* pointer to next ring in q_vector */
328	void *desc;			/* Descriptor ring memory */
329	struct device *dev;		/* Used for DMA mapping */
330	struct net_device *netdev;	/* netdev ring maps to */
331	struct bpf_prog *xdp_prog;
332	union {
333		struct i40e_tx_buffer *tx_bi;
334		struct i40e_rx_buffer *rx_bi;
335		struct xdp_buff **rx_bi_zc;
336	};
337	DECLARE_BITMAP(state, __I40E_RING_STATE_NBITS);
338	u16 queue_index;		/* Queue number of ring */
339	u8 dcb_tc;			/* Traffic class of ring */
340	u8 __iomem *tail;
341
342	/* Storing xdp_buff on ring helps in saving the state of partially built
343	 * packet when i40e_clean_rx_ring_irq() must return before it sees EOP
344	 * and to resume packet building for this ring in the next call to
345	 * i40e_clean_rx_ring_irq().
346	 */
347	struct xdp_buff xdp;
348
349	/* Next descriptor to be processed; next_to_clean is updated only on
350	 * processing EOP descriptor
351	 */
352	u16 next_to_process;
353	/* high bit set means dynamic, use accessor routines to read/write.
354	 * hardware only supports 2us resolution for the ITR registers.
355	 * these values always store the USER setting, and must be converted
356	 * before programming to a register.
357	 */
358	u16 itr_setting;
359
360	u16 count;			/* Number of descriptors */
361	u16 reg_idx;			/* HW register index of the ring */
362	u16 rx_buf_len;
363
364	/* used in interrupt processing */
365	u16 next_to_use;
366	u16 next_to_clean;
367	u16 xdp_tx_active;
368
369	u8 atr_sample_rate;
370	u8 atr_count;
371
372	bool ring_active;		/* is ring online or not */
373	bool arm_wb;		/* do something to arm write back */
374	u8 packet_stride;
375
376	u16 flags;
377#define I40E_TXR_FLAGS_WB_ON_ITR		BIT(0)
378#define I40E_RXR_FLAGS_BUILD_SKB_ENABLED	BIT(1)
379#define I40E_TXR_FLAGS_XDP			BIT(2)
380
381	/* stats structs */
382	struct i40e_queue_stats	stats;
383	struct u64_stats_sync syncp;
384	union {
385		struct i40e_tx_queue_stats tx_stats;
386		struct i40e_rx_queue_stats rx_stats;
387	};
388
389	unsigned int size;		/* length of descriptor ring in bytes */
390	dma_addr_t dma;			/* physical address of ring */
391
392	struct i40e_vsi *vsi;		/* Backreference to associated VSI */
393	struct i40e_q_vector *q_vector;	/* Backreference to associated vector */
394
395	struct rcu_head rcu;		/* to avoid race on free */
396	u16 next_to_alloc;
 
 
 
 
 
 
 
 
397
398	struct i40e_channel *ch;
399	u16 rx_offset;
400	struct xdp_rxq_info xdp_rxq;
401	struct xsk_buff_pool *xsk_pool;
 
402} ____cacheline_internodealigned_in_smp;
403
404static inline bool ring_uses_build_skb(struct i40e_ring *ring)
405{
406	return !!(ring->flags & I40E_RXR_FLAGS_BUILD_SKB_ENABLED);
407}
408
409static inline void set_ring_build_skb_enabled(struct i40e_ring *ring)
410{
411	ring->flags |= I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
412}
413
414static inline void clear_ring_build_skb_enabled(struct i40e_ring *ring)
415{
416	ring->flags &= ~I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
417}
418
419static inline bool ring_is_xdp(struct i40e_ring *ring)
420{
421	return !!(ring->flags & I40E_TXR_FLAGS_XDP);
422}
423
424static inline void set_ring_xdp(struct i40e_ring *ring)
425{
426	ring->flags |= I40E_TXR_FLAGS_XDP;
427}
428
429#define I40E_ITR_ADAPTIVE_MIN_INC	0x0002
430#define I40E_ITR_ADAPTIVE_MIN_USECS	0x0002
431#define I40E_ITR_ADAPTIVE_MAX_USECS	0x007e
432#define I40E_ITR_ADAPTIVE_LATENCY	0x8000
433#define I40E_ITR_ADAPTIVE_BULK		0x0000
 
434
435struct i40e_ring_container {
436	struct i40e_ring *ring;		/* pointer to linked list of ring(s) */
437	unsigned long next_update;	/* jiffies value of next update */
438	unsigned int total_bytes;	/* total bytes processed this int */
439	unsigned int total_packets;	/* total packets processed this int */
440	u16 count;
441	u16 target_itr;			/* target ITR setting for ring(s) */
442	u16 current_itr;		/* current ITR setting for ring(s) */
443};
444
445/* iterator for handling rings in ring container */
446#define i40e_for_each_ring(pos, head) \
447	for (pos = (head).ring; pos != NULL; pos = pos->next)
448
449static inline unsigned int i40e_rx_pg_order(struct i40e_ring *ring)
450{
451#if (PAGE_SIZE < 8192)
452	if (ring->rx_buf_len > (PAGE_SIZE / 2))
453		return 1;
454#endif
455	return 0;
456}
457
458#define i40e_rx_pg_size(_ring) (PAGE_SIZE << i40e_rx_pg_order(_ring))
459
460bool i40e_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
461netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
462u16 i40e_lan_select_queue(struct net_device *netdev, struct sk_buff *skb,
463			  struct net_device *sb_dev);
464void i40e_clean_tx_ring(struct i40e_ring *tx_ring);
465void i40e_clean_rx_ring(struct i40e_ring *rx_ring);
466int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring);
467int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring);
468void i40e_free_tx_resources(struct i40e_ring *tx_ring);
469void i40e_free_rx_resources(struct i40e_ring *rx_ring);
470int i40e_napi_poll(struct napi_struct *napi, int budget);
471void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
472u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw);
473void i40e_detect_recover_hung(struct i40e_vsi *vsi);
474int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
475bool __i40e_chk_linearize(struct sk_buff *skb);
476int i40e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
477		  u32 flags);
478bool i40e_is_non_eop(struct i40e_ring *rx_ring,
479		     union i40e_rx_desc *rx_desc);
480
481/**
482 * i40e_get_head - Retrieve head from head writeback
483 * @tx_ring:  tx ring to fetch head of
484 *
485 * Returns value of Tx ring head based on value stored
486 * in head write-back location
487 **/
488static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
489{
490	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
491
492	return le32_to_cpu(*(volatile __le32 *)head);
493}
494
495/**
496 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
497 * @skb:     send buffer
 
498 *
499 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
500 * there is not enough descriptors available in this ring since we need at least
501 * one descriptor.
502 **/
503static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
504{
505	const skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
506	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
507	int count = 0, size = skb_headlen(skb);
508
509	for (;;) {
510		count += i40e_txd_use_count(size);
511
512		if (!nr_frags--)
513			break;
514
515		size = skb_frag_size(frag++);
516	}
517
518	return count;
519}
520
521/**
522 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
523 * @tx_ring: the ring to be checked
524 * @size:    the size buffer we want to assure is available
525 *
526 * Returns 0 if stop is not needed
527 **/
528static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
529{
530	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
531		return 0;
532	return __i40e_maybe_stop_tx(tx_ring, size);
533}
534
535/**
536 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
537 * @skb:      send buffer
538 * @count:    number of buffers used
539 *
540 * Note: Our HW can't scatter-gather more than 8 fragments to build
541 * a packet on the wire and so we need to figure out the cases where we
542 * need to linearize the skb.
543 **/
544static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
545{
546	/* Both TSO and single send will work if count is less than 8 */
547	if (likely(count < I40E_MAX_BUFFER_TXD))
548		return false;
549
550	if (skb_is_gso(skb))
551		return __i40e_chk_linearize(skb);
552
553	/* we can support up to 8 data buffers for a single send */
554	return count != I40E_MAX_BUFFER_TXD;
555}
556
557/**
558 * txring_txq - Find the netdev Tx ring based on the i40e Tx ring
559 * @ring: Tx ring to find the netdev equivalent of
560 **/
561static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
562{
563	return netdev_get_tx_queue(ring->netdev, ring->queue_index);
564}
565#endif /* _I40E_TXRX_H_ */
v5.4
  1/* SPDX-License-Identifier: GPL-2.0 */
  2/* Copyright(c) 2013 - 2018 Intel Corporation. */
  3
  4#ifndef _I40E_TXRX_H_
  5#define _I40E_TXRX_H_
  6
  7#include <net/xdp.h>
 
  8
  9/* Interrupt Throttling and Rate Limiting Goodies */
 10#define I40E_DEFAULT_IRQ_WORK      256
 11
 12/* The datasheet for the X710 and XL710 indicate that the maximum value for
 13 * the ITR is 8160usec which is then called out as 0xFF0 with a 2usec
 14 * resolution. 8160 is 0x1FE0 when written out in hex. So instead of storing
 15 * the register value which is divided by 2 lets use the actual values and
 16 * avoid an excessive amount of translation.
 17 */
 18#define I40E_ITR_DYNAMIC	0x8000	/* use top bit as a flag */
 19#define I40E_ITR_MASK		0x1FFE	/* mask for ITR register value */
 20#define I40E_MIN_ITR		     2	/* reg uses 2 usec resolution */
 21#define I40E_ITR_100K		    10	/* all values below must be even */
 22#define I40E_ITR_50K		    20
 23#define I40E_ITR_20K		    50
 24#define I40E_ITR_18K		    60
 25#define I40E_ITR_8K		   122
 26#define I40E_MAX_ITR		  8160	/* maximum value as per datasheet */
 27#define ITR_TO_REG(setting) ((setting) & ~I40E_ITR_DYNAMIC)
 28#define ITR_REG_ALIGN(setting) __ALIGN_MASK(setting, ~I40E_ITR_MASK)
 29#define ITR_IS_DYNAMIC(setting) (!!((setting) & I40E_ITR_DYNAMIC))
 30
 31#define I40E_ITR_RX_DEF		(I40E_ITR_20K | I40E_ITR_DYNAMIC)
 32#define I40E_ITR_TX_DEF		(I40E_ITR_20K | I40E_ITR_DYNAMIC)
 33
 34/* 0x40 is the enable bit for interrupt rate limiting, and must be set if
 35 * the value of the rate limit is non-zero
 36 */
 37#define INTRL_ENA                  BIT(6)
 38#define I40E_MAX_INTRL             0x3B    /* reg uses 4 usec resolution */
 39#define INTRL_REG_TO_USEC(intrl) ((intrl & ~INTRL_ENA) << 2)
 40
 41/**
 42 * i40e_intrl_usec_to_reg - convert interrupt rate limit to register
 43 * @intrl: interrupt rate limit to convert
 44 *
 45 * This function converts a decimal interrupt rate limit to the appropriate
 46 * register format expected by the firmware when setting interrupt rate limit.
 47 */
 48static inline u16 i40e_intrl_usec_to_reg(int intrl)
 49{
 50	if (intrl >> 2)
 51		return ((intrl >> 2) | INTRL_ENA);
 52	else
 53		return 0;
 54}
 55#define I40E_INTRL_8K              125     /* 8000 ints/sec */
 56#define I40E_INTRL_62K             16      /* 62500 ints/sec */
 57#define I40E_INTRL_83K             12      /* 83333 ints/sec */
 58
 59#define I40E_QUEUE_END_OF_LIST 0x7FF
 60
 61/* this enum matches hardware bits and is meant to be used by DYN_CTLN
 62 * registers and QINT registers or more generally anywhere in the manual
 63 * mentioning ITR_INDX, ITR_NONE cannot be used as an index 'n' into any
 64 * register but instead is a special value meaning "don't update" ITR0/1/2.
 65 */
 66enum i40e_dyn_idx_t {
 67	I40E_IDX_ITR0 = 0,
 68	I40E_IDX_ITR1 = 1,
 69	I40E_IDX_ITR2 = 2,
 70	I40E_ITR_NONE = 3	/* ITR_NONE must not be used as an index */
 71};
 72
 73/* these are indexes into ITRN registers */
 74#define I40E_RX_ITR    I40E_IDX_ITR0
 75#define I40E_TX_ITR    I40E_IDX_ITR1
 76#define I40E_PE_ITR    I40E_IDX_ITR2
 77
 78/* Supported RSS offloads */
 79#define I40E_DEFAULT_RSS_HENA ( \
 80	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_UDP) | \
 81	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_SCTP) | \
 82	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP) | \
 83	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_OTHER) | \
 84	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV4) | \
 85	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_UDP) | \
 86	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP) | \
 87	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_SCTP) | \
 88	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_OTHER) | \
 89	BIT_ULL(I40E_FILTER_PCTYPE_FRAG_IPV6) | \
 90	BIT_ULL(I40E_FILTER_PCTYPE_L2_PAYLOAD))
 91
 92#define I40E_DEFAULT_RSS_HENA_EXPANDED (I40E_DEFAULT_RSS_HENA | \
 93	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV4_TCP_SYN_NO_ACK) | \
 94	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV4_UDP) | \
 95	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV4_UDP) | \
 96	BIT_ULL(I40E_FILTER_PCTYPE_NONF_IPV6_TCP_SYN_NO_ACK) | \
 97	BIT_ULL(I40E_FILTER_PCTYPE_NONF_UNICAST_IPV6_UDP) | \
 98	BIT_ULL(I40E_FILTER_PCTYPE_NONF_MULTICAST_IPV6_UDP))
 99
100#define i40e_pf_get_default_rss_hena(pf) \
101	(((pf)->hw_features & I40E_HW_MULTIPLE_TCP_UDP_RSS_PCTYPE) ? \
102	  I40E_DEFAULT_RSS_HENA_EXPANDED : I40E_DEFAULT_RSS_HENA)
103
104/* Supported Rx Buffer Sizes (a multiple of 128) */
105#define I40E_RXBUFFER_256   256
106#define I40E_RXBUFFER_1536  1536  /* 128B aligned standard Ethernet frame */
107#define I40E_RXBUFFER_2048  2048
108#define I40E_RXBUFFER_3072  3072  /* Used for large frames w/ padding */
109#define I40E_MAX_RXBUFFER   9728  /* largest size for single descriptor */
110
111/* NOTE: netdev_alloc_skb reserves up to 64 bytes, NET_IP_ALIGN means we
112 * reserve 2 more, and skb_shared_info adds an additional 384 bytes more,
113 * this adds up to 512 bytes of extra data meaning the smallest allocation
114 * we could have is 1K.
115 * i.e. RXBUFFER_256 --> 960 byte skb (size-1024 slab)
116 * i.e. RXBUFFER_512 --> 1216 byte skb (size-2048 slab)
117 */
118#define I40E_RX_HDR_SIZE I40E_RXBUFFER_256
119#define I40E_PACKET_HDR_PAD (ETH_HLEN + ETH_FCS_LEN + (VLAN_HLEN * 2))
120#define i40e_rx_desc i40e_32byte_rx_desc
121
122#define I40E_RX_DMA_ATTR \
123	(DMA_ATTR_SKIP_CPU_SYNC | DMA_ATTR_WEAK_ORDERING)
124
125/* Attempt to maximize the headroom available for incoming frames.  We
126 * use a 2K buffer for receives and need 1536/1534 to store the data for
127 * the frame.  This leaves us with 512 bytes of room.  From that we need
128 * to deduct the space needed for the shared info and the padding needed
129 * to IP align the frame.
130 *
131 * Note: For cache line sizes 256 or larger this value is going to end
132 *	 up negative.  In these cases we should fall back to the legacy
133 *	 receive path.
134 */
135#if (PAGE_SIZE < 8192)
136#define I40E_2K_TOO_SMALL_WITH_PADDING \
137((NET_SKB_PAD + I40E_RXBUFFER_1536) > SKB_WITH_OVERHEAD(I40E_RXBUFFER_2048))
138
139static inline int i40e_compute_pad(int rx_buf_len)
140{
141	int page_size, pad_size;
142
143	page_size = ALIGN(rx_buf_len, PAGE_SIZE / 2);
144	pad_size = SKB_WITH_OVERHEAD(page_size) - rx_buf_len;
145
146	return pad_size;
147}
148
149static inline int i40e_skb_pad(void)
150{
151	int rx_buf_len;
152
153	/* If a 2K buffer cannot handle a standard Ethernet frame then
154	 * optimize padding for a 3K buffer instead of a 1.5K buffer.
155	 *
156	 * For a 3K buffer we need to add enough padding to allow for
157	 * tailroom due to NET_IP_ALIGN possibly shifting us out of
158	 * cache-line alignment.
159	 */
160	if (I40E_2K_TOO_SMALL_WITH_PADDING)
161		rx_buf_len = I40E_RXBUFFER_3072 + SKB_DATA_ALIGN(NET_IP_ALIGN);
162	else
163		rx_buf_len = I40E_RXBUFFER_1536;
164
165	/* if needed make room for NET_IP_ALIGN */
166	rx_buf_len -= NET_IP_ALIGN;
167
168	return i40e_compute_pad(rx_buf_len);
169}
170
171#define I40E_SKB_PAD i40e_skb_pad()
172#else
173#define I40E_2K_TOO_SMALL_WITH_PADDING false
174#define I40E_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
175#endif
176
177/**
178 * i40e_test_staterr - tests bits in Rx descriptor status and error fields
179 * @rx_desc: pointer to receive descriptor (in le64 format)
180 * @stat_err_bits: value to mask
181 *
182 * This function does some fast chicanery in order to return the
183 * value of the mask which is really only used for boolean tests.
184 * The status_error_len doesn't need to be shifted because it begins
185 * at offset zero.
186 */
187static inline bool i40e_test_staterr(union i40e_rx_desc *rx_desc,
188				     const u64 stat_err_bits)
189{
190	return !!(rx_desc->wb.qword1.status_error_len &
191		  cpu_to_le64(stat_err_bits));
192}
193
194/* How many Rx Buffers do we bundle into one write to the hardware ? */
195#define I40E_RX_BUFFER_WRITE	32	/* Must be power of 2 */
196#define I40E_RX_INCREMENT(r, i) \
197	do {					\
198		(i)++;				\
199		if ((i) == (r)->count)		\
200			i = 0;			\
201		r->next_to_clean = i;		\
202	} while (0)
203
204#define I40E_RX_NEXT_DESC(r, i, n)		\
205	do {					\
206		(i)++;				\
207		if ((i) == (r)->count)		\
208			i = 0;			\
209		(n) = I40E_RX_DESC((r), (i));	\
210	} while (0)
211
212#define I40E_RX_NEXT_DESC_PREFETCH(r, i, n)		\
213	do {						\
214		I40E_RX_NEXT_DESC((r), (i), (n));	\
215		prefetch((n));				\
216	} while (0)
217
218#define I40E_MAX_BUFFER_TXD	8
219#define I40E_MIN_TX_LEN		17
220
221/* The size limit for a transmit buffer in a descriptor is (16K - 1).
222 * In order to align with the read requests we will align the value to
223 * the nearest 4K which represents our maximum read request size.
224 */
225#define I40E_MAX_READ_REQ_SIZE		4096
226#define I40E_MAX_DATA_PER_TXD		(16 * 1024 - 1)
227#define I40E_MAX_DATA_PER_TXD_ALIGNED \
228	(I40E_MAX_DATA_PER_TXD & ~(I40E_MAX_READ_REQ_SIZE - 1))
229
230/**
231 * i40e_txd_use_count  - estimate the number of descriptors needed for Tx
232 * @size: transmit request size in bytes
233 *
234 * Due to hardware alignment restrictions (4K alignment), we need to
235 * assume that we can have no more than 12K of data per descriptor, even
236 * though each descriptor can take up to 16K - 1 bytes of aligned memory.
237 * Thus, we need to divide by 12K. But division is slow! Instead,
238 * we decompose the operation into shifts and one relatively cheap
239 * multiply operation.
240 *
241 * To divide by 12K, we first divide by 4K, then divide by 3:
242 *     To divide by 4K, shift right by 12 bits
243 *     To divide by 3, multiply by 85, then divide by 256
244 *     (Divide by 256 is done by shifting right by 8 bits)
245 * Finally, we add one to round up. Because 256 isn't an exact multiple of
246 * 3, we'll underestimate near each multiple of 12K. This is actually more
247 * accurate as we have 4K - 1 of wiggle room that we can fit into the last
248 * segment.  For our purposes this is accurate out to 1M which is orders of
249 * magnitude greater than our largest possible GSO size.
250 *
251 * This would then be implemented as:
252 *     return (((size >> 12) * 85) >> 8) + 1;
253 *
254 * Since multiplication and division are commutative, we can reorder
255 * operations into:
256 *     return ((size * 85) >> 20) + 1;
257 */
258static inline unsigned int i40e_txd_use_count(unsigned int size)
259{
260	return ((size * 85) >> 20) + 1;
261}
262
263/* Tx Descriptors needed, worst case */
264#define DESC_NEEDED (MAX_SKB_FRAGS + 6)
265#define I40E_MIN_DESC_PENDING	4
266
267#define I40E_TX_FLAGS_HW_VLAN		BIT(1)
268#define I40E_TX_FLAGS_SW_VLAN		BIT(2)
269#define I40E_TX_FLAGS_TSO		BIT(3)
270#define I40E_TX_FLAGS_IPV4		BIT(4)
271#define I40E_TX_FLAGS_IPV6		BIT(5)
272#define I40E_TX_FLAGS_FCCRC		BIT(6)
273#define I40E_TX_FLAGS_FSO		BIT(7)
274#define I40E_TX_FLAGS_TSYN		BIT(8)
275#define I40E_TX_FLAGS_FD_SB		BIT(9)
276#define I40E_TX_FLAGS_UDP_TUNNEL	BIT(10)
277#define I40E_TX_FLAGS_VLAN_MASK		0xffff0000
278#define I40E_TX_FLAGS_VLAN_PRIO_MASK	0xe0000000
279#define I40E_TX_FLAGS_VLAN_PRIO_SHIFT	29
280#define I40E_TX_FLAGS_VLAN_SHIFT	16
281
282struct i40e_tx_buffer {
283	struct i40e_tx_desc *next_to_watch;
284	union {
285		struct xdp_frame *xdpf;
286		struct sk_buff *skb;
287		void *raw_buf;
288	};
289	unsigned int bytecount;
290	unsigned short gso_segs;
291
292	DEFINE_DMA_UNMAP_ADDR(dma);
293	DEFINE_DMA_UNMAP_LEN(len);
294	u32 tx_flags;
295};
296
297struct i40e_rx_buffer {
298	dma_addr_t dma;
299	union {
300		struct {
301			struct page *page;
302			__u32 page_offset;
303			__u16 pagecnt_bias;
304		};
305		struct {
306			void *addr;
307			u64 handle;
308		};
309	};
310};
311
312struct i40e_queue_stats {
313	u64 packets;
314	u64 bytes;
315};
316
317struct i40e_tx_queue_stats {
318	u64 restart_queue;
319	u64 tx_busy;
320	u64 tx_done_old;
321	u64 tx_linearize;
322	u64 tx_force_wb;
 
323	int prev_pkt_ctr;
324};
325
326struct i40e_rx_queue_stats {
327	u64 non_eop_descs;
328	u64 alloc_page_failed;
329	u64 alloc_buff_failed;
330	u64 page_reuse_count;
331	u64 realloc_count;
 
 
332};
333
334enum i40e_ring_state_t {
335	__I40E_TX_FDIR_INIT_DONE,
336	__I40E_TX_XPS_INIT_DONE,
337	__I40E_RING_STATE_NBITS /* must be last */
338};
339
340/* some useful defines for virtchannel interface, which
341 * is the only remaining user of header split
342 */
343#define I40E_RX_DTYPE_NO_SPLIT      0
344#define I40E_RX_DTYPE_HEADER_SPLIT  1
345#define I40E_RX_DTYPE_SPLIT_ALWAYS  2
346#define I40E_RX_SPLIT_L2      0x1
347#define I40E_RX_SPLIT_IP      0x2
348#define I40E_RX_SPLIT_TCP_UDP 0x4
349#define I40E_RX_SPLIT_SCTP    0x8
350
351/* struct that defines a descriptor ring, associated with a VSI */
352struct i40e_ring {
353	struct i40e_ring *next;		/* pointer to next ring in q_vector */
354	void *desc;			/* Descriptor ring memory */
355	struct device *dev;		/* Used for DMA mapping */
356	struct net_device *netdev;	/* netdev ring maps to */
357	struct bpf_prog *xdp_prog;
358	union {
359		struct i40e_tx_buffer *tx_bi;
360		struct i40e_rx_buffer *rx_bi;
 
361	};
362	DECLARE_BITMAP(state, __I40E_RING_STATE_NBITS);
363	u16 queue_index;		/* Queue number of ring */
364	u8 dcb_tc;			/* Traffic class of ring */
365	u8 __iomem *tail;
366
 
 
 
 
 
 
 
 
 
 
 
367	/* high bit set means dynamic, use accessor routines to read/write.
368	 * hardware only supports 2us resolution for the ITR registers.
369	 * these values always store the USER setting, and must be converted
370	 * before programming to a register.
371	 */
372	u16 itr_setting;
373
374	u16 count;			/* Number of descriptors */
375	u16 reg_idx;			/* HW register index of the ring */
376	u16 rx_buf_len;
377
378	/* used in interrupt processing */
379	u16 next_to_use;
380	u16 next_to_clean;
 
381
382	u8 atr_sample_rate;
383	u8 atr_count;
384
385	bool ring_active;		/* is ring online or not */
386	bool arm_wb;		/* do something to arm write back */
387	u8 packet_stride;
388
389	u16 flags;
390#define I40E_TXR_FLAGS_WB_ON_ITR		BIT(0)
391#define I40E_RXR_FLAGS_BUILD_SKB_ENABLED	BIT(1)
392#define I40E_TXR_FLAGS_XDP			BIT(2)
393
394	/* stats structs */
395	struct i40e_queue_stats	stats;
396	struct u64_stats_sync syncp;
397	union {
398		struct i40e_tx_queue_stats tx_stats;
399		struct i40e_rx_queue_stats rx_stats;
400	};
401
402	unsigned int size;		/* length of descriptor ring in bytes */
403	dma_addr_t dma;			/* physical address of ring */
404
405	struct i40e_vsi *vsi;		/* Backreference to associated VSI */
406	struct i40e_q_vector *q_vector;	/* Backreference to associated vector */
407
408	struct rcu_head rcu;		/* to avoid race on free */
409	u16 next_to_alloc;
410	struct sk_buff *skb;		/* When i40e_clean_rx_ring_irq() must
411					 * return before it sees the EOP for
412					 * the current packet, we save that skb
413					 * here and resume receiving this
414					 * packet the next time
415					 * i40e_clean_rx_ring_irq() is called
416					 * for this ring.
417					 */
418
419	struct i40e_channel *ch;
 
420	struct xdp_rxq_info xdp_rxq;
421	struct xdp_umem *xsk_umem;
422	struct zero_copy_allocator zca; /* ZC allocator anchor */
423} ____cacheline_internodealigned_in_smp;
424
425static inline bool ring_uses_build_skb(struct i40e_ring *ring)
426{
427	return !!(ring->flags & I40E_RXR_FLAGS_BUILD_SKB_ENABLED);
428}
429
430static inline void set_ring_build_skb_enabled(struct i40e_ring *ring)
431{
432	ring->flags |= I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
433}
434
435static inline void clear_ring_build_skb_enabled(struct i40e_ring *ring)
436{
437	ring->flags &= ~I40E_RXR_FLAGS_BUILD_SKB_ENABLED;
438}
439
440static inline bool ring_is_xdp(struct i40e_ring *ring)
441{
442	return !!(ring->flags & I40E_TXR_FLAGS_XDP);
443}
444
445static inline void set_ring_xdp(struct i40e_ring *ring)
446{
447	ring->flags |= I40E_TXR_FLAGS_XDP;
448}
449
450#define I40E_ITR_ADAPTIVE_MIN_INC	0x0002
451#define I40E_ITR_ADAPTIVE_MIN_USECS	0x0002
452#define I40E_ITR_ADAPTIVE_MAX_USECS	0x007e
453#define I40E_ITR_ADAPTIVE_LATENCY	0x8000
454#define I40E_ITR_ADAPTIVE_BULK		0x0000
455#define ITR_IS_BULK(x) (!((x) & I40E_ITR_ADAPTIVE_LATENCY))
456
457struct i40e_ring_container {
458	struct i40e_ring *ring;		/* pointer to linked list of ring(s) */
459	unsigned long next_update;	/* jiffies value of next update */
460	unsigned int total_bytes;	/* total bytes processed this int */
461	unsigned int total_packets;	/* total packets processed this int */
462	u16 count;
463	u16 target_itr;			/* target ITR setting for ring(s) */
464	u16 current_itr;		/* current ITR setting for ring(s) */
465};
466
467/* iterator for handling rings in ring container */
468#define i40e_for_each_ring(pos, head) \
469	for (pos = (head).ring; pos != NULL; pos = pos->next)
470
471static inline unsigned int i40e_rx_pg_order(struct i40e_ring *ring)
472{
473#if (PAGE_SIZE < 8192)
474	if (ring->rx_buf_len > (PAGE_SIZE / 2))
475		return 1;
476#endif
477	return 0;
478}
479
480#define i40e_rx_pg_size(_ring) (PAGE_SIZE << i40e_rx_pg_order(_ring))
481
482bool i40e_alloc_rx_buffers(struct i40e_ring *rxr, u16 cleaned_count);
483netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
 
 
484void i40e_clean_tx_ring(struct i40e_ring *tx_ring);
485void i40e_clean_rx_ring(struct i40e_ring *rx_ring);
486int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring);
487int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring);
488void i40e_free_tx_resources(struct i40e_ring *tx_ring);
489void i40e_free_rx_resources(struct i40e_ring *rx_ring);
490int i40e_napi_poll(struct napi_struct *napi, int budget);
491void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector);
492u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw);
493void i40e_detect_recover_hung(struct i40e_vsi *vsi);
494int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size);
495bool __i40e_chk_linearize(struct sk_buff *skb);
496int i40e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
497		  u32 flags);
 
 
498
499/**
500 * i40e_get_head - Retrieve head from head writeback
501 * @tx_ring:  tx ring to fetch head of
502 *
503 * Returns value of Tx ring head based on value stored
504 * in head write-back location
505 **/
506static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
507{
508	void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
509
510	return le32_to_cpu(*(volatile __le32 *)head);
511}
512
513/**
514 * i40e_xmit_descriptor_count - calculate number of Tx descriptors needed
515 * @skb:     send buffer
516 * @tx_ring: ring to send buffer on
517 *
518 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
519 * there is not enough descriptors available in this ring since we need at least
520 * one descriptor.
521 **/
522static inline int i40e_xmit_descriptor_count(struct sk_buff *skb)
523{
524	const skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
525	unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
526	int count = 0, size = skb_headlen(skb);
527
528	for (;;) {
529		count += i40e_txd_use_count(size);
530
531		if (!nr_frags--)
532			break;
533
534		size = skb_frag_size(frag++);
535	}
536
537	return count;
538}
539
540/**
541 * i40e_maybe_stop_tx - 1st level check for Tx stop conditions
542 * @tx_ring: the ring to be checked
543 * @size:    the size buffer we want to assure is available
544 *
545 * Returns 0 if stop is not needed
546 **/
547static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
548{
549	if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
550		return 0;
551	return __i40e_maybe_stop_tx(tx_ring, size);
552}
553
554/**
555 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
556 * @skb:      send buffer
557 * @count:    number of buffers used
558 *
559 * Note: Our HW can't scatter-gather more than 8 fragments to build
560 * a packet on the wire and so we need to figure out the cases where we
561 * need to linearize the skb.
562 **/
563static inline bool i40e_chk_linearize(struct sk_buff *skb, int count)
564{
565	/* Both TSO and single send will work if count is less than 8 */
566	if (likely(count < I40E_MAX_BUFFER_TXD))
567		return false;
568
569	if (skb_is_gso(skb))
570		return __i40e_chk_linearize(skb);
571
572	/* we can support up to 8 data buffers for a single send */
573	return count != I40E_MAX_BUFFER_TXD;
574}
575
576/**
577 * txring_txq - Find the netdev Tx ring based on the i40e Tx ring
578 * @ring: Tx ring to find the netdev equivalent of
579 **/
580static inline struct netdev_queue *txring_txq(const struct i40e_ring *ring)
581{
582	return netdev_get_tx_queue(ring->netdev, ring->queue_index);
583}
584#endif /* _I40E_TXRX_H_ */